SERENE ER Analysis Part 1-SERENE CD: Exposure-Response Analysis of Higher Versus Standard Adalimumab Dosing Regimens for Patients With Moderately to Severely Active Crohn Disease.

SERENE CD (NCT02065570) evaluated whether a higher adalimumab induction dose would improve patients with Crohn disease response and suggested a flat dose-response relationship for efficacy in the induction study. We investigated exposure-response relationships in induction and maintenance studies considering patients' baseline characteristics. Adalimumab exposures were simulated using the established population pharmacokinetic model. Efficacy end points (clinical remission/endoscopic response) at Weeks 4, 12, and 56 were evaluated in exposure-response analyses using multivariable logistic regression. Analyses showed an increasing trend with heterogeneity between induction regimens, which suggested that average concentration has an impact on coprimary efficacy end points within each group, but data did not fit a single-response curve. Although higher concentrations within arms were associated with improved outcomes, increasing the concentration through a higher induction dose was not associated with increasing clinical remission/endoscopic response at Week 4/12. A model including inverse effective clearance eliminated heterogeneity and described trends across induction regimens with a single curve. In the maintenance study, the response rates at Week 56 showed no heterogeneity. In the induction study, patients with lower effective adalimumab clearance responded better, whereas in the maintenance study average concentration drove primary efficacy end points at Week 56. Research extending these findings to other indications is needed.

SERENE ER Analysis Part 2 SERENE-UC: Exposure-response Analysis of Higher Versus Standard Adalimumab Dosing Regimens for Patients with Moderately to Severely Active Ulcerative Colitis.

SERENE UC (NCT02065622) evaluated whether a higher adalimumab induction regimen improved patients with ulcerative colitis (UC) response, but a flat dose-response relationship was found in the induction study. We investigated exposure-response (ER) relationships in induction and maintenance studies considering patients' baseline characteristics. Adalimumab exposures were simulated using the established population pharmacokinetic model. Multivariable logistic regressions were used to assess the efficacy endpoints (clinical remission, endoscopic remission, endoscopic improvement) at weeks 8 and 52. In the induction study, an increasing ER trend with heterogeneity between induction regimens was shown, suggesting average concentration (Cavg) had a significant impact on primary efficacy endpoints within each group. However, data were not described by a single ER curve. Using inverse effective clearance as the exposure metric described trends across induction regimens with a single curve. Patients with inherently lower effective adalimumab clearance responded better. The patient response rates at week 52 showed no heterogeneity. A short-term increase in adalimumab dose did not drive better responses for induction, and apparent ER relationships were better explained by patient-inherent lower clearance. Conversely, during maintenance up to week 52, increasing the concentration via dose translated to better responses more robustly. The ER findings for SERENE UC were consistent with SERENE CD.

Integrating machine learning with pharmacokinetic models: Benefits of scientific machine learning in adding neural networks components to existing PK models.

Recently, the use of machine-learning (ML) models for pharmacokinetic (PK) modeling has grown significantly. Although most of the current approaches use ML techniques as black boxes, there are only a few that have proposed interpretable architectures which integrate mechanistic knowledge. In this work, we use as the test case a one-compartment PK model using a scientific machine learning (SciML) framework and consider learning an unknown absorption using neural networks, while simultaneously estimating other parameters of drug distribution and elimination. We generate simulated data with different sampling strategies to show that our model can accurately predict concentrations in extrapolation tasks, including new dosing regimens with different sparsity levels, and produce reliable forecasts even for new patients. By using a scenario of fitting PK data with complex absorption, we demonstrate that including known physiological structure into an SciML model allows us to obtain highly accurate predictions while preserving the interpretability of classical compartmental models.

Assessing the Impact of Immunogenicity and Improving Prediction of Trough Concentrations: Population Pharmacokinetic Modeling of Adalimumab in Patients with Crohn's Disease and Ulcerative Colitis.

BACKGROUND AND OBJECTIVE: Predicting adalimumab pharmacokinetics (PK) for patients impacted by anti-drug antibodies (ADA) has been challenging. The present study assessed the performance of the adalimumab immunogenicity assays in predicting which patients with Crohn's disease (CD) and ulcerative colitis (UC) have low adalimumab trough concentrations; and aimed to improve predictive performance of adalimumab population PK (popPK) model in CD and UC patients whose PK was impacted by ADA. METHODS: Adalimumab PK and immunogenicity data obtained from 1459 patients in SERENE CD (NCT02065570) and SERENE UC (NCT02065622) were analyzed. Adalimumab immunogenicity was assessed using electrochemiluminescence (ECL) and enzyme-linked immunosorbent (ELISA) assays. From these assays, three analytical approaches (ELISA concentrations, titer, and signal-to-noise [S/N] measurements) were tested as predictors for classifying patients with/without low concentrations potentially affected by immunogenicity. The performance of different thresholds for these analytical procedures was assessed using receiver operating characteristic curves and precision-recall curves. Based on the results from the most sensitive immunogenicity analytical procedure, patients were classified into PK-not-ADA-impacted and PK-ADA-impacted subpopulations. Stepwise popPK modeling was implemented to fit the PK data to an empirical adalimumab two-compartment model with linear elimination and ADA delay compartments to account for the time delay to generate ADA. Model performance was assessed by visual predictive checks and goodness-of-fit plots. RESULTS: The classical ELISA-based classification (with 20 ng/mL ADA as lower threshold) showed a good balance of precision and recall, to determine which patients had at least 30% adalimumab concentrations below 1 µg/mL. Titer-based classification with the lower limit of quantitation (LLOQ) as threshold showed higher sensitivity to classify these patients compared to the ELISA-based approach. Therefore, patients were classified as PK-ADA-impacted or PK-not-ADA impacted using the LLOQ titer threshold. In the stepwise modeling approach ADA-independent parameters were first fit using PK data from titer-PK-not-ADA-impacted population. The identified ADA-independent covariates included the effect of indication, weight, baseline fecal calprotectin, baseline C-reactive protein, baseline albumin on clearance; and sex and weight on volume of distribution of the central compartment. Pharmacokinetic-ADA-driven dynamics were characterized using PK data for the PK-ADA-impacted population. The categorical covariate based on the ELISA classification was the best at describing the additional effect of immunogenicity analytical approaches on ADA synthesis rate. The model was able to adequately describe the central tendency and variability for PK-ADA-impacted CD/UC patients. CONCLUSIONS: The ELISA assay was found to be optimal for capturing impact of ADA on PK. The developed adalimumab popPK model is robust in predicting PK profiles for CD and UC patients whose PK was impacted by ADA.

Upadacitinib Population Pharmacokinetics and Exposure-Response Relationships in Ulcerative Colitis Patients.

BACKGROUND AND OBJECTIVE: Upadacitinib, an oral selective and reversible Janus kinase (JAK) inhibitor, showed favorable efficacy and safety in patients with moderate-to-severe ulcerative colitis (UC). The objective was to characterize upadacitinib pharmacokinetics in UC patients across Phase 2b and 3 trials and evaluate the relationships between upadacitinib plasma exposures and key efficacy or safety endpoints. METHODS: Population pharmacokinetics and exposure-response analyses were performed to characterize upadacitinib pharmacokinetics in UC patients and evaluate the relationships between plasma exposures and key efficacy or safety endpoints at the end of 8-week induction and 52-week maintenance periods. Data from 1234 UC patients from Phase 2 and 3 induction trials and 449 UC patients from a Phase 3 maintenance trial were used for these analyses. Additionally, data from patients with rheumatoid arthritis, atopic dermatitis, Crohn's disease, and healthy volunteers were used in the pharmacokinetics analysis. Quartile plots and logistic regression models were used to evaluate the exposure-response relationships across upadacitinib doses of 7.5-45 mg once daily (QD) for induction and 15-30 mg QD for maintenance. RESULTS: Upadacitinib plasma exposures were dose-proportional in UC patients across the evaluated dose range. Upadacitinib pharmacokinetics in UC were consistent between the induction and maintenance periods, and with other patient populations. Upadacitinib plasma exposures associated with the 45 mg QD induction dose maximized efficacy for Week 8 clinical and endoscopic endpoints. Plasma exposures associated with upadacitinib 30 mg maintenance dose provided additional incremental benefit compared to 15 mg QD for Week 52 key clinical and endoscopic endpoints. No trends were observed in the evaluated safety events with increasing plasma exposures at the end of induction or maintenance periods. CONCLUSION: These analyses supported selection of upadacitinib UC induction and maintenance doses. TRIAL REGISTRATION: Data from studies NCT02819635 and NCT03653026 were included in these analyses.

Age Structure Can Account for Delayed Logistic Proliferation of Scratch Assays.

Scratch assays are in vitro methods for studying cell migration. In these experiments, a scratch is made on a cell monolayer and recolonisation of the scratched region is imaged to quantify cell migration rates. Typically, scratch assays are modelled by reaction diffusion equations depicting cell migration by Fickian diffusion and proliferation by a logistic term. In a recent paper (Jin et al. in Bull Math Biol 79(5):1028-1050, 2017), the authors observed experimentally that during the early stage of the recolonisation process, there is a disturbance phase where proliferation is not logistic, and this is followed by a growth phase where proliferation appears to be logistic. The authors did not identify the precise mechanism that causes the disturbance phase but showed that ignoring it can lead to incorrect parameter estimates. The aim of this work is to show that a nonlinear age-structured population model can account for the two phases of proliferation in scratch assays. The model consists of an age-structured cell cycle model of a cell population, coupled with an ordinary differential equation describing the resource concentration dynamics in the substrate. The model assumes a resource-dependent cell cycle threshold age, above which cells are able to proliferate. By studying the dynamics of the full system in terms of the subpopulations of cells that can proliferate and the ones that can not, we are able to find conditions under which the model captures the two-phase behaviour. Through numerical simulations, we are able to show that the interplay between the resource concentration in the substrate and the cell subpopulations dynamics can explain the biphasic dynamics.

A stochastic model for tumour control probability that accounts for repair from sublethal damage.

The tumour control probability (TCP) is the probability that a treatment regimen of radiation therapy (RT) eradicates all tumour cells in a given tissue. To decrease the toxic effects on healthy cells, RT is usually delivered over a period of weeks in a series of fractions. This allows tumour cells to repair sublethal damage (RSD) caused by radiation. In this article, we introduce a stochastic model for tumour response to radiotherapy which accounts for the effects of RSD. The tumour is subdivided into two cell types: 'affected' cells which have been damaged by RT and 'unaffected' cells which have not. The model is formulated as a birth-death process for which we can derive an explicit formula for the TCP. We apply our model to prostate cancer, and find that the radiosensitivity parameters and the probability of sublethal damage during radiation are the parameters to which the TCP predictions are most sensitive. We compare our TCP predictions to those given by Zaider and Minerbo's one-class model (Zaider & Minerbo, 2000) and Dawson and Hillen's two-class model (Dawson & Hillen, 2006) and find that for low doses of radiation, our model predicts a lower TCP. Finally, we find that when the probability of sublethal damage during radiation is large, the mean field assumption overestimates the TCP.